In reference to calculating adjacency matrix from gradient of image, I found something in python.
large-adjacency-matrix-from-image-in-python
I want to calculate an adjacency matrix based on 4 or 8 neighboring pixels. I also found http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3408910/
How can i do this with 4 or 8 neighbors? I want to do this in C++. I already have gradient image for use.
For the sake of simplicity, assume that the gradient image is a square pixel bitmap of size n x n. Assign an ordinal number to each pixel by row-major counting starting in the northwestern corner.
Define the (n^2 x n^2) adjacency matrix A = (a_ij)_i,j=1..n^2 as follows:
a_i(i-n) = 1; i > n // northern neighbour
a_i(i+1) = 1; (i-1) mod n < n-1 // eastern neighbour
a_i(i-1) = 1; (i-1) mod n > 0 // western neighbour
a_i(i+n) = 1; i <= n^2 - n // southern neighbour
a_ij = 0; else
For 8 neighbours per pixel add
a_i(i-n+1) = 1; i > n and (i-n-1) mod n < n-1 // northeastern neighbour
a_i(i-n-1) = 1; i > n and (i-n-1) mod n > 0 // northwestern neighbour
a_i(i+n+1) = 1; i <= n^2 - n and (i+n-1) mod n < n-1 // southeastern neighbour
a_i(i+n-1) = 1; i <= n^2 - n and (i+n-1) mod n > 0 // southwestern neighbour
Instead of 1 you may assign the weights calculated from the gradient between adjacent pixels. Note that 0 entries would change to M, M representing a sufficiently large number ( infinite, since the respective cells are no neighbours, but that requires the implementation take special provisions ).
A will be sparse and will have a regular structure, for efficiency you should probably employ a class for sparse matrix processing. This SO question provides some suggestions.
Related
I attempted this SPOJ problem.
Problem:
AMR10J - Mixing Chemicals
There are N bottles each having a different chemical. For each chemical i, you have determined C[i] which means that mixing chemicals i and C[i] causes an explosion. You have K distinct boxes. In how many ways can you divide the N chemicals into those boxes such that no two chemicals in the same box can cause an explosion together?
INPUT
The first line of input is the number of test cases T. T test cases follow each containing 2 lines.
The first line of each test case contains 2 integers N and K.
The second line of each test case contains N integers, the ith integer denoting the value C[i]. The chemicals are numbered from 0 to N-1.
OUTPUT
For each testcase, output the number of ways modulo 1,000,000,007.
CONSTRAINTS
T <= 50
2 <= N <= 100
2 <= K <= 1000
0 <= C[i] < N
For all i, i != C[i]
SAMPLE INPUT
3
3 3
1 2 0
4 3
1 2 0 0
3 2
1 2 0
SAMPLE OUTPUT
6
12
0
EXPLANATION
In the first test case, we cannot mix any 2 chemicals. Hence, each of the 3 boxes must contain 1 chemical, which leads to 6 ways in total.
In the third test case, we cannot put the 3 chemicals in the 2 boxes satisfying all the 3 conditions.
The summary of the problem, given a set of chemicals and a set of boxes, count how many possible ways to place these chemicals in boxes such that no chemicals will explode.
At first I used brute force method to solve the problem, I recursively place chemicals in boxes and count valid configurations, I got TLE at my first attempt.
Later I learned that the problem can be solved with graph colouring.
I can represent chemicals as vertexes and there'a an edge between chemicals if they cannot be placed each other.
And the set of boxes can be used as vertex colours, all I need to do was to count how many different valid colourings of the graph.
I applyed this concept to solve the problem unfortunately I got TLE again. I don't know how to improve my code, I need help.
code:
#include <bits/stdc++.h>
#define MAXN 100
using namespace std;
const int mod = (int) 1e9 + 7;
int n;
int k;
int ways;
void greedy_coloring(vector<int> adj[], int color[])
{
int u = 0;
for (; u < n; ++u)
if (color[u] == -1)//found first uncolored vertex
break;
if (u == n)//no uncolored vertexex means all vertexes are colored
{
ways = (ways + 1) % mod;
return;
}
bool available[k];
memset(available, true, sizeof(available));
for (int v : adj[u])
if (color[v] != -1)//if the adjacent vertex colored, make its color unavailable
available[color[v]] = false;
for (int c = 0; c < k; ++c)
if (available[c])
{
color[u] = c;
greedy_coloring(adj, color);
color[u] = -1;//don't forgot to reset the color
}
}
int main()
{
ios_base::sync_with_stdio(false);
cin.tie(NULL);
int T;
cin >> T;
while (T--)
{
cin >> n >> k;
vector<int> adj[n];
int c[n];
for (int i = 0; i < n; ++i)
{
cin >> c[i];
adj[i].push_back(c[i]);
adj[c[i]].push_back(i);
}
ways = 0;
int color[n];
memset(color, -1, sizeof(color));
greedy_coloring(adj, color);
cout << ways << "\n";
}
return 0;
}
Counting the number of colorings in a general graph is #P-hard, but this graph has some special structure, which I'll exploit in a minute after I enumerate some basic properties of counting colorings. The first observation is that, if the graph has a node with no neighbors, if we delete that node, the number of colorings decreases by a factor of k. The second observation is that, if a node has exactly one neighbor and we delete it, the number of colorings decreases by a factor of k-1. The third is that the number of colorings is equal to the product of the number of colorings for each connected component. The fourth is that we can delete all but one parallel edge.
Using these properties, it suffices to determine a formula for each connected component of the 2-core of this graph, which is a simple cycle of some length. Let P(n) and C(n) be the number of ways to color a path or cycle respectively with n nodes. We use the basic properties above to find
P(n) = k (k-1)^(n-1).
Finding a formula for C(n) I think requires the deletion contraction formula, which leads to a recurrence
C(3) = k (k-1) (k-2), i.e., three nodes of different colors;
C(n) = P(n) - C(n-1) = k (k-1)^(n-1) - C(n-1).
Multiply the above recurrence by (-1)^n.
(-1)^3 C(3) = -k (k-1) (k-2)
(-1)^n C(n) = (-1)^n k (k-1)^(n-1) - (-1)^n C(n-1)
= (-1)^n k (k-1)^(n-1) + (-1)^(n-1) C(n-1)
(-1)^n C(n) - (-1)^(n-1) C(n-1) = (-1)^n k (k-1)^(n-1)
Let D(n) = (-1)^n C(n).
D(3) = -k (k-1) (k-2)
D(n) - D(n-1) = (-1)^n k (k-1)^(n-1)
Now we can write D(n) as a telescoping sum:
D(n) = [sum_{i=4}^n (D(n) - D(n-1))] + D(3)
D(n) = [sum_{i=4}^n (-1)^n k (k-1)^(n-1)] - k (k-1) (k-2).
Break it down as two geometric sums which then cancel nicely.
D(n) = [sum_{i=4}^n (-1)^n ((k-1) + 1) (k-1)^(n-1)] - k (k-1) (k-2)
= sum_{i=4}^n (1-k)^n - sum_{i=4}^n (1-k)^(n-1) - k (k-1) (k-2)
= (1-k)^n - (1-k)^3 - k (k-1) (k-2)
= (1-k)^n - (1 - 3k + 3k^2 - k^3) - (2k - 3k^2 + k^3)
= (1-k)^n - (1-k)
C(n) = (-1)^n (1-k)^n - (-1)^n (1-k)
= (k-1)^n + (-1)^n (k-1).
Note that after removing all parallel edges, we can have at most n edges. This means that in any one connected component we can only see one cycle (and simple at that), which makes the combinatorics rather straightforward. (Cycles are only dependent on how many edges each node can spawn, which is capped at 1.)
Second example:
k = 3
<< 0 <-- 3
/ ^
/ ^
1 --> 2
Since cycles are self contained, any connection to one removes the possibility of another. In the example above, we cannot make a second cycle involving node 3 by adding more nodes, and the same issue would extend to any subsequent connected nodes.
It should be enough, therefore, to perform a search, separating out connected components and marking their node count and whether they contain a cycle. Given a connected component, where c of the nodes are part of a cycle and m nodes are not, we have the following formula (David Eisenstat helped me correct my combinatoric for the count of colourings of a cycle):
if the component has a cycle:
[(k - 1)^c + (-1)^c * (k - 1)] *
(k - 1)^(m)
otherwise:
k * (k - 1)^(m - 1)
As David Eisenstat noted, multiply all these results for the final tally.
I am writing this question fishing for any state-of-the-art software or methods that can quickly compute the intersection of N 2D polygons (the convex hulls of projected convex polyhedrons), and M 2D polygons where typically N >> M. N may be in the order or at least 1M polygons and N in the order 50k. I've searched for some time now, but I keep coming up with the same answer shown below.
Use boost and a loop to
compute the projection of the polyhedron (not the bottleneck)
compute the convex hull of said polyhedron (bottleneck)
compute the intersection of the projected polyhedron and existing 2D polygon (major bottleneck).
This loop is repeated NK times where typically K << M, and K is the average number of 2D polygons intersecting a single projected polyhedron. This is done to reduce the number of computations.
The problem with this is that if I have N=262144 and M=19456 it takes about 129 seconds (when multithreaded by polyhedron), and this must be done about 300 times. Ideally, I would like to reduce the computation time to about 1 second for the above sizes, so I was wondering if someone could help point to some software or literature that could improve efficiency.
[EDIT]
#sehe's request I'm posting the most relevant parts of the code. I haven't compiled it, so this is just to get the gist... this code assumes, there are voxels and pixels, but the shapes can be anything. The order of the points in the grid can be any, but the indices of where the points reside in the grid are the same.
#include <boost/geometry/geometry.hpp>
#include <boost/geometry/geometries/point.hpp>
#include <boost/geometry/geometries/ring.hpp>
const std::size_t Dimension = 2;
typedef boost::geometry::model::point<float, Dimension, boost::geometry::cs::cartesian> point_2d;
typedef boost::geometry::model::polygon<point_2d, false /* is cw */, true /* closed */> polygon_2d;
typedef boost::geometry::model::box<point_2d> box_2d;
std::vector<float> getOverlaps(std::vector<float> & projected_grid_vx, // projected voxels
std::vector<float> & pixel_grid_vx, // pixels
std::vector<int> & projected_grid_m, // number of voxels in each dimension
std::vector<int> & pixel_grid_m, // number of pixels in each dimension
std::vector<float> & pixel_grid_omega, // size of the pixel grid in cm
int projected_grid_size, // total number of voxels
int pixel_grid_size) { // total number of pixels
std::vector<float> overlaps(projected_grid_size * pixel_grid_size);
std::vector<float> h(pixel_grid_m.size());
for(int d=0; d < pixel_grid_m.size(); d++) {
h[d] = (pixel_grid_omega[2*d+1] - pixel_grid_omega[2*d]) / pixel_grid_m[d];
}
for(int i=0; i < projected_grid_size; i++){
std::vector<float> point_indices(8);
point_indices[0] = i;
point_indices[1] = i + 1;
point_indices[2] = i + projected_grid_m[0];
point_indices[3] = i + projected_grid_m[0] + 1;
point_indices[4] = i + projected_grid_m[0] * projected_grid_m[1];
point_indices[5] = i + projected_grid_m[0] * projected_grid_m[1] + 1;
point_indices[6] = i + (projected_grid_m[1] + 1) * projected_grid_m[0];
point_indices[7] = i + (projected_grid_m[1] + 1) * projected_grid_m[0] + 1;
std::vector<float> vx_corners(8 * projected_grid_m.size());
for(int vn = 0; vn < 8; vn++) {
for(int d = 0; d < projected_grid_m.size(); d++) {
vx_corners[vn + d * 8] = projected_grid_vx[point_indices[vn] + d * projeted_grid_size];
}
}
polygon_2d proj_voxel;
for(int vn = 0; vn < 8; vn++) {
point_2d poly_pt(vx_corners[2 * vn], vx_corners[2 * vn + 1]);
boost::geometry::append(proj_voxel, poly_pt);
}
boost::geometry::correct(proj_voxel);
polygon_2d proj_voxel_hull;
boost::geometry::convex_hull(proj_voxel, proj_voxel_hull);
box_2d bb_proj_vox;
boost::geometry::envelope(proj_voxel_hull, bb_proj_vox);
point_2d min_pt = bb_proj_vox.min_corner();
point_2d max_pt = bb_proj_vox.max_corner();
// then get min and max indices of intersecting bins
std::vector<float> min_idx(projected_grid_m.size() - 1),
max_idx(projected_grid_m.size() - 1);
// compute min and max indices of incidence on the pixel grid
// this is easy assuming you have a regular grid of pixels
min_idx[0] = std::min( (float) std::max( std::floor((min_pt.get<0>() - pixel_grid_omega[0]) / h[0] - 0.5 ), 0.), pixel_grid_m[0]-1);
min_idx[1] = std::min( (float) std::max( std::floor((min_pt.get<1>() - pixel_grid_omega[2]) / h[1] - 0.5 ), 0.), pixel_grid_m[1]-1);
max_idx[0] = std::min( (float) std::max( std::floor((max_pt.get<0>() - pixel_grid_omega[0]) / h[0] + 0.5 ), 0.), pixel_grid__m[0]-1);
max_idx[1] = std::min( (float) std::max( std::floor((max_pt.get<1>() - pixel_grid_omega[2]) / h[1] + 0.5 ), 0.), pixel_grid_m[1]-1);
// iterate only over pixels which intersect the projected voxel
for(int iy = min_idx[1]; iy <= max_idx[1]; iy++) {
for(int ix = min_idx[0]; ix <= max_idx[0]; ix++) {
int idx = ix + iy * pixel_grid_size[0]; // `first' index of pixel corner point
polygon_2d pix_poly;
for(int pn = 0; pn < 4; pn++) {
point_2d pix_corner_pt(
pixel_grid_vx[idx + pn % 2 + (pn / 2) * pixel_grid_m[0]],
pixel_grid_vx[idx + pn % 2 + (pn / 2) * pixel_grid_m[0] + pixel_grid_size]
);
boost::geometry::append(pix_poly, pix_corner_pt);
}
boost::geometry::correct( pix_poly );
//make this into a convex hull since the order of the point may be any
polygon_2d pix_hull;
boost::geometry::convex_hull(pix_poly, pix_hull);
// on to perform intersection
std::vector<polygon_2d> vox_pix_ints;
polygon_2d vox_pix_int;
try {
boost::geometry::intersection(proj_voxel_hull, pix_hull, vox_pix_ints);
} catch ( std::exception e ) {
// skip since these may coincide at a point or line
continue;
}
// both are convex so only one intersection expected
vox_pix_int = vox_pix_ints[0];
overlaps[i + idx * projected_grid_size] = boost::geometry::area(vox_pix_int);
}
} // end intersection for
} //end projected_voxel for
return overlaps;
}
You could create the ratio of polygon to bounding box:
This could be done computationally once to arrive at an avgerage poly area to BB ratio R constant.
Or you could do it with geometry using a circle bounded by its BB Since your using only projected polyhedron:
R = 0.0;
count = 0;
for (each poly) {
count++;
R += polyArea / itsBoundingBoxArea;
}
R = R/count;
Then calculate the summation of intersection of bounding boxes.
Sbb = 0.0;
for (box1, box2 where box1.isIntersecting(box2)) {
Sbb += box1.intersect(box2);
}
Then:
Approximation = R * Sbb
All of this would not work if concave polys were allowed. Because a concave poly can occupy less than 1% of it's bounding box. You will still have to find the convex hull.
Alternatively, If you can find the polygons area quicker than its hull, you could use the actual computed average poly area. This would give you a decent approximation as well while avoiding both poly intersection and wrapping.
Hm, the problem seems similar to doing "collision-detection" i game-engines. Or "potentially visible sets".
While I don't know much about the current state-of-the-art, i remember an optimization was to enclose objects in spheres, since checking overlaps between spheres (or circles in 2D) is really cheap.
In order to speed-up checks for collisions, objects were often put into search-structures (e.g. a sphere-tree (circle-tree in 2D case)). Basically organizing the space into a hierarchical structure, to make queries for overlaps fast.
So basically my suggestion boils down to: Try looking at algorithms for collision-detection i game-engines.
Assumption
I'm assuming that you mean "intersections" and not intersection. Moreover, It is not the expected use case that most of the individual polys from M and N will overlap at the same time. If this assumption is true then:
Answer
The way this is done with 2D game engines is by having a scene graph where every object has a bounding box. Then place all the the polygons into a node in an quadtree according to their location determined by bounding box. Then the task becomes parallel because each node can be processed separately for intersection.
Here is the wiki for quadtree:
Quadtree Wiki
An octree could be used when in 3D.
It actually doesn't even have to be a octree. You could get the same results with any space partition. You could find the maximum separation of polys (lets call it S). And create say S/10 space partitions. Then you would have 10 separate spaces to execute in parallel. Not only would it be concurrent, but It would no longer be M * N time since not every poly must be compared against every other poly.
I need to filter given width of lines in a image.
I am coding a program which will detect lines of road image. And I found something like that but can't understand logic of it. My function has to do that:
I will send image and width of line in terms of pixel size(e.g 30 pixel width), the function will filter just these lines in image.
I found that code:
void filterWidth(Mat image, int tau) // tau=width of line I want to filter
int aux = 0;
for (int j = 0; j < quad.rows; ++j)
{
unsigned char *ptRowSrc = quad.ptr<uchar>(j);
unsigned char *ptRowDst = quadDst.ptr<uchar>(j);
for (int i = tau; i < quad.cols - tau; ++i)
{
if (ptRowSrc[i] != 0)
{
aux = 2 * ptRowSrc[i];
aux += -ptRowSrc[i - tau];
aux += -ptRowSrc[i + tau];
aux += -abs((int)(ptRowSrc[i - tau] - ptRowSrc[i + tau]));
aux = (aux < 0) ? (0) : (aux);
aux = (aux > 255) ? (255) : (aux);
ptRowDst[i] = (unsigned char)aux;
}
}
}
What is the mathematical explanation of that code? And how does that work?
Read up about convolution filters. This code is a particular case of a 1 dimensional convolution filter (it only convolves with other pixels on the currently processed line).
The value of aux is started with 2 * the current pixel value, then pixels on either side of it at distance tau are being subtracted from that value. Next the absolute difference of those two pixels is also subtracted from it. Finally it is capped to the range 0...255 before being stored in the output image.
If you have an image:
0011100
This convolution will cause the centre 1 to gain the value:
2 * 1
- 0
- 0
- abs(0 - 0)
= 2
The first '1' will become:
2 * 1
- 0
- 1
- abs(0 - 1)
= 0
And so will the third '1' (it's a mirror image).
And of course the 0 values will always stay zero or become negative, which will be capped back to 0.
This is a rather weird filter. It takes the pixel values three by three on the same line, with a tau spacing. Let these values by Vl, V and Vr.
The filter computes - Vl + 2 V - Vr, which can be seen as a second derivative, and deducts |Vl - Vr|, which can be seen as a first derivative (also called gradient). The second derivative gives a maximum response in case of a maximum configuration (Vl < V > Vr); the first derivative gives a minimum response in case of a symmetric configuration (Vl = Vr).
So the global filter will give a maximum response for a symmetric maximum (like with a light road on a dark background, vertical, with a width less than 2.tau).
By rearranging the terms, you can see that the filter also yields the smallest of the left and right gradients, V - Vm and V - Vp (clamped to zero).
I'm working on an OpenCL project to generate very large hermitian (symmetric) matrices, and I am trying to determine the best way to generate the work IDs.
A hermitian matrix is symmetric along the diagonal, so that M(i,j) = M*(j,i).
In the brute force way, the for loop looks like:
for(int i = 0; i < N; i++)
{
for(int j = 0; j < N; j++)
{
complex<float> result = doSomeCalculation();
M(i,j) = result;
}
}
However, taking advantage of the hermitian property, the loop can be made to be twice as efficient by only calculating the upper triangular part of the matrix and duplicating the result in the lower triangular part:
for(int i = 0; i < N; i++)
{
for(int j = i; j < N; j++)
{
complex<float> result = doSomeCalculation();
M(i,j) = result;
M(j,i) = conj(result);
}
}
In both loops, doSomeCalculation() is an expensive operation, and each entry in the matrix is completely uncorrelated from every other entry (i.e. the problem is stupidly parallel).
My question is this:
How can I implement the second loop with doSomeCalculation as an OpenCL kernel so that the thread IDs are most efficiently used (i.e. so that the thread calculates both M(i,j) and M(j,i) without having to call doSomeCalculation() twice)?
You need to use a linear index, for example you can index every element of your matrix in this way:
0 1 2 ... N-1
* N-2 ... 2N-2
....
* * 2N-1 ... N(N+1)/2 -1
That is, the index K is given by:
k=iN-i*(i+1)/2+j
Where N is the size of the matrix and (i,j) are respectively the 0-based indices of the row and the column.
This relationship can be inverted; see the answer of this question, which I report here for completeness:
i = floor( ( 2*N+1 - sqrt( (2N+1)*(2N+1) - 8*k ) ) / 2 ) ;
j = k - N*i + i*(i+1)/2 ;
So you need to enqueue a 1D kernel with N(N+1)/2 work items, and you can decide by yourself the size of the workgroup (usually 64 items per work group is a good choice).
Then in the OpenCL code you can retrieve the index K by using:
int k = get_group_id(0)*64 + get_local_id(0);
And then use the two relationships above the index of the matrix element you need to compute.
Moreover, notice that you can also save space by representing your hermitian matrix as a linear vector with N(N+1)/2 elements.
If your matrices are really big, than you can dice up your NxN matrix into (N/k)x(N/k) tiles, each of size kxk. As soon as you need only a half of the data, you create 1D NDRange of size local_group_size * (N/k)x(N/k)/2 roughly.
Every tile of matrix is processed by one LocalGroup (size of LocalGroup is of your choice). The idea is that you create an array on Host side, which contain position of every WorkGroup in matrix. Kernel stub should look like follows:
void __kernel myKernel(
__global int* coords,
....)
{
int2 WorkGroupPositionInMatrix = vload2(get_group_id(0), coords);
...
DoCalculation();
...
WriteResultTwice();
...
return;
}
What you need to do by hand - is to cope with thouse WorkGroups, which will be placed on the matrix diagonal. If matrix size is big, than overhead for LocalGroups, placed on diagonal is negligible.
A right triangle can be cut in half vertically and the smaller portion rotated to fit with the larger portion to form a rectangle of equal area. Therefore it is easy to make your triangular global work area into one that is rectangular, which fits OpenCL.
See my answer here: OpenCL efficient way to group a lower triangular matrix
I create a minimum distance filter for points.
The function takes a stream of points (x1,y1,x2,y2...) and removes the corresponding ones.
void minDistanceFilter(vector<float> &points, float distance = 0.0)
{
float p0x, p0y;
float dx, dy, dsq;
float mdsq = distance*distance; // minimum distance square
unsigned i, j, n = points.size();
for(i=0; i<n; ++i)
{
p0x = points[i];
p0y = points[i+1];
for(j=0; j<n; j+=2)
{
//if (i == j) continue; // discard itself (seems like it slows down the algorithm)
dx = p0x - points[j]; // delta x (p0x - p1x)
dy = p0y - points[j+1]; // delta y (p0y - p1y)
dsq = dx*dx + dy*dy; // distance square
if (dsq < mdsq)
{
auto del = points.begin() + j;
points.erase(del,del+3);
n = points.size(); // update n
j -= 2; // decrement j
}
}
}
}
The only problem that is very slow, due to it tests all points against all points (n^2).
How could it be improved?
kd-trees or range trees could be used for your problem. However, if you want to code from scratch and want something simpler, then you can use a hash table structure. For each point (a,b), hash using the key (round(a/d),round(b/d)) and store all the points that have the same key in a list. Then, for each key (m,n) in your hash table, compare all points in the list to the list of points that have key (m',n') for all 9 choices of (m',n') where m' = m + (-1 or 0 or 1) and n' = n + (-1 or 0 or 1). These are the only points that can be within distance d of your points that have key (m,n). The downside compared to a kd-tree or range tree is that for a given point, you are effectively searching within a square of side length 3*d for points that might have distance d or less, instead of searching within a square of side length 2*d which is what you would get if you used a kd-tree or range tree. But if you are coding from scratch, this is easier to code; also kd-trees and range trees are kinda overkill if you only have one universal distance d that you care about for all points.
Look up range tree, e.g. en.wikipedia.org/wiki/Range_tree . You can use this structure to store 2-dimensional points and very quickly find all the points that lie inside a query rectangle. Since you want to find points within a certain distance d of a point (a,b), your query rectangle will need to be [a-d,a+d]x[b-d,b+d] and then you test any points found inside the rectangle to make sure they are actually within distance d of (a,b). Range tree can be built in O(n log n) time and space, and range queries take O(log n + k) time where k is the number of points found in the rectangle. Seems optimal for your problem.